如何在没有 SDK 的情况下调试 OpenAI API 调用

When working with the OpenAI API, issues may arise such as incorrect responses, the model ignoring system prompts, or being rate limited without understanding why. Common debugging methods include adding print statements throughout the code, setting up LangSmith or Helicone, or staring at the OpenAI dashboard usage page. However, these methods can be inefficient for quick debugging. A useful trick involves utilizing environment variables, as many OpenAI client libraries allow overriding the base URL. The official Python client uses OPENAI_BASE_URL, which is also respected by LangChain and nearly every wrapper. By pointing this environment variable to an inspector that forwards to the real API, debugging becomes more manageable.